Reader Leader — An AI reading tutor that respects Irish accents

What it actually does

It listens to every word — then decides what to do next.

Speech-to-text only hears the words. The hard part is the judgement: help now, or stay quiet? Reader Leader makes three decisions, word by word, and then writes a reading record for the teacher.

01

Prompt

Step in with a gentle hint when the child needs support — enough to teach, little enough to leave the reading flowing.

02

Model

Say the word aloud as a calm demonstration when a hint isn't enough. No red marks. No buzzers. Just the word.

03

Stay silent

Recognise a dialect variant and let the child read on. The hardest decision, and the most important. When confidence is low, it stays quiet — silence costs less than a knock to confidence.

The demo moment

Why this matters

Children learn to read by reading aloud to someone.

It's the best-evidenced fluency intervention we have. But it's rationed by the one resource a school can't buy more of: adult attention. Three jobs, one teacher, thirty children, one voice.

Mentor — guide the child through the text.
Motivate — keep them engaged and confident.
Measure — track progress word by word.

Teachers can't hear every child, every day. Reader Leader does the listening nobody has time for — so every child gets a listener, not just a few.

#2Ireland's place in the PIRLS 2021 international reading rankings.A strong national average — but no room for complacency.

A national average hides very different experiences.

Some children are being left behind:

DEIS schoolsChildren with dyslexiaNeuro-divergent learnersLate learners & EAL pupils

What makes us different

Most read-aloud tools were built for American voices. We're built for Irish and British ones.

The incumbents are proven at scale and procurement-ready. But four things make a school in Ireland or the UK hesitate — and each one is where we focus.

1

North-American English only

Accent, vocabulary, spelling, content and culture are built for a different classroom.

2

False corrections

An accent mismatch tells a child they're wrong when they're not — and they stop reading aloud.

3

Closed content libraries

Teachers can't bring their own texts, so schools lose pedagogical control.

4

Unpublished intervention policy

Schools can't see how the AI decides to act. Reader Leader publishes its approach.

We license

The ears.

We don't rebuild speech recognition. We license it and can swap vendors, partnering with providers who have real children's-voice data. They build the ears.

+

We build

The judgement & the Irish accent layer.

Our own asset is a hand-built list of valid Irish and British pronunciations for each word. If a child matches one, it isn't an error. That's the hard, local work — and it's ours.

We measure wrong corrections by accent group — and we publish them, never averaged. As far as we can find, nobody else in this category does. Accuracy in one accent, done properly, beats weak coverage everywhere.

How it works

One reading session. Ten to fifteen minutes. Self-recording.

It comes after phonics — exactly where the DfE Reading Framework puts fluency. Each session fits a single school reading slot, and writes its own running record.

1
Read it. Lead it.
The child chooses a story and reads it aloud in their own voice. Levels are friendly guides, not tests.
2
Listen, word by word
Reader Leader listens to each word and decides: prompt, model, or stay silent.
3
Calm feedback
A gentle hint, then the word if needed. No red marks, no buzzers — we keep the child reading, not worrying.
4
A running record, done for you
The session writes a reading record automatically — progress, errors and difficult words, by child and across the class.

🍎

For teachers

You stay in charge. Every child gets a listener, not just a few. Running records are done for you, you see who's struggling and where — and you can mark any flagged word as correct. Teachers have the final say.

📚

For parents

Follow your child's progress and practise together at home, between school sessions. Patient, private practice — the kind that suits a struggling reader.

🧭

For educators & leaders

It supports the teacher's programme, it doesn't replace it. Fluency is the bridge from decoding to understanding — and it aligns with where the DfE Reading Framework with the NCCA alignment to be finalised.

Where we are, honestly

We'd rather you heard the second list from us than found it yourselves.

We're at prototype stage. Here's what's built and demonstrable today — and what isn't yet. We won't invent a number we haven't measured.

✓ Built & demonstrable

Read-aloud, end-to-end
Prompt / model / stay-silent, live
Five-way error classifier
One accent measured
Running record generated
A false-correction baseline

○ Not yet

Multi-accent coverage
Production ASR licensing
A full content library
The full teacher dashboard
Measured learning gain
Finalised data-governance policies

The metric that matters

Why false correction, never averaged.

A child told they're wrong when they're not stops reading aloud. The two failure modes aren't equal — and this is the number that decides whether the product is usable at all. It's our headline promise, and we won't guess it.

Catastrophic

False correction

Confidence and trust both drop. One observed false correction can end adoption.

Recoverable

Missed error

One lost teaching moment — costly, but recoverable in the next session.

“We don't know yet. Here's exactly how we'll find out.” We're building an accent-balanced test set now, and we'll publish our first false-correction figures by accent group.

The model

Flexible licences and great value

Planned pricing. Schools pay from their literacy or digital-learning budget; trusts pay centrally; tutors and parents pay themselves. One school licence costs less than the classroom hours it gives back in a single term.

School licence

Annual, priced per pupil, banded by school size.

€5–€20per pupil, per year

Group or trust licence

One contract across many schools, with group-level reporting.

€5,000+per site, per year

Tutor seat

One professional running records across their own caseload.

€9–€25per month

Parent subscription

One child, at home, between school sessions.

€19–€29per month

Progress reporting

Anonymised, aggregated reading data for trusts and funders.

€500+per report

Indicative planned pricing, not a live price list. Final pricing follows the pilots.

Data protection & the EU AI Act

The school stays in control of the data.

At this competition stage we use only synthetic audio or recordings made by adult team members — no real child has used the system. Before any pilot with real children, we design conservatively to the higher standard.

A DPIA is a hard pre-pilot gate

A Data Protection Impact Assessment will be completed with the data controller, safeguards implemented and the outcome formally signed off before any real child data is processed — and kept as a living process, not a one-off form.

Child audio isn't used to train the model

Not by default and not during the proposed pilot. Any future use for training would need a separate legal, ethical and technical assessment — and would never be a condition of accessing the service.

No emotion inference

We measure observable reading events — timing, pauses, word accuracy. We don't use voice or biometric data to infer a child's emotions, stress or confidence. Emotion inference is expressly excluded from our intended use.

Designed to the higher standard

The EU AI Act may classify systems that evaluate learning as high-risk. Until classification is confirmed, we assume nothing is exempt: documented risk management, data governance, traceable logs and meaningful teacher oversight.

Data minimisation & EU hosting

Our design preference is hosting within the EU/EEA, retaining audio only for the shortest justifiable period. We won't state a retention period or location as fact until it's assessed, agreed and documented.

Teacher oversight, always

No unreviewed AI judgement should decide an educational outcome for a child. Teachers can review, correct, reject or override any AI output — and that feedback improves our accent list.

Testing in Ireland will be guided by the BERA Ethical Guidelines, the National Policy Statement on Ensuring Research Integrity, Children First and Tusla safeguarding guidance, and the Data Protection Commission's guidance on children's data.

What happens next

Leave your email — we're picking pilot schools.

You can't buy Reader Leader today. But we're choosing pilot schools now, and we'll bring you in early. We're looking for:

Pilot schools — three or four DEIS primaries with mixed accents, plus a tutor practice, this school year.
A speech partner — an ASR provider with real children's-voice data.
Introductions — literacy leads of UK academy trusts, where one conversation reaches many schools.

An AI reading tutor that knows when to stay quiet.